DeepStruc: towards structure solution from pair distribution function data using deep generative models.

Area of Science:

• Materials Science
• Computational Materials Science
• Nanotechnology

Background:

• Determining the structure of nanostructured materials with limited long-range order is a significant challenge in materials development.
• Traditional methods often struggle with the inherent disorder and complexity of nanomaterials.

Purpose of the Study:

• To introduce DeepStruc, a novel deep learning algorithm for solving nanoparticle structures directly from Pair Distribution Function (PDF) data.
• To demonstrate the capability of DeepStruc in solving structures from both simulated and experimental PDF data, including those outside the training set.

Main Methods:

• Utilized a conditional variational autoencoder (a type of deep learning model) to develop the DeepStruc algorithm.
• Applied DeepStruc to analyze PDFs from various monometallic nanoparticle structures.
• Tested DeepStruc on simulated and experimental PDF data, including complex systems with hexagonal close-packed (hcp), face-centered cubic (fcc), and stacking-faulted nanoparticles.

Main Results:

• DeepStruc successfully solved simple monometallic nanoparticle structures directly from PDF data.
• The algorithm accurately determined structures from both simulated and experimental PDFs, even for nanoparticles not included in the training data.
• DeepStruc identified stacking-faulted nanoparticles as an intermediate state between hcp and fcc structures, successfully solving their structures from PDFs.

Conclusions:

• DeepStruc represents a significant advancement in the direct structure solution of nanomaterials from scattering data.
• The algorithm shows promise as a general approach for tackling the structural complexity of nanomaterials.
• This work paves the way for more efficient characterization and development of novel nanostructured materials.